AB-620 career: agent builder roles
Passing AB-620 earns the Microsoft Certified: AI Agent Builder Associate credential, and Microsoft tags it with two roles on its certification page: App Maker and Developer. The audience profile is more specific than either tag. It describes a “professional developer or advanced builder” who typically works as an IT application developer, a consultant or an independent software vendor (ISV) partner. No job title called “AI Agent Builder” is promised by the credential; it describes a body of work, not a position.
Makers versus developers: where AB-620 sits
Microsoft’s tagging puts this exam on the border between two populations that are usually kept apart. A citizen maker builds a first Copilot Studio agent with a knowledge source and a few topics. A developer wires agents to systems through code, protocols and pipelines. AB-620 deliberately targets the second group while staying inside Copilot Studio.
The profile’s wording makes the split concrete. It expects hands-on experience with “basic knowledge sources, instructions, tools, and topics” as the starting point, and then lists what the working role adds:
- integrating agents with Microsoft Foundry, Model Context Protocol (MCP) servers, custom connectors, APIs and Microsoft Fabric;
- automating tasks with computer use;
- building multi-agent solutions and agents grounded in enterprise sources “such as ServiceNow, SAP, and others”.
A maker who has only built a single-topic FAQ agent has not yet done most of that list. A developer who has never opened Copilot Studio has done none of it in this product. The credential speaks to people who have done the second list inside the first tool.
Who sits next to you: the collaborators Microsoft names
The audience profile ends with a list of colleagues the role works with. It reads like a staffing diagram for an agent programme:
| Counterpart named by Microsoft | What the AB-620 holder typically needs from them |
|---|---|
| Microsoft 365 administrators and Copilot administrators | Tenant settings, channel availability, who may publish agents |
| Power Platform administrators and Copilot Studio administrators | Environments, data policies, pipelines |
| Foundry administrators | Model access and agents that Copilot Studio hands tasks to |
| Copilot Studio agent builders | Makers who build the simpler agents you extend or review |
| Agentic AI business solutions architects and Copilot Studio architects | The design you implement, or the role you may grow into |
Two lessons follow. The job is cross-functional, so interview preparation should include how decisions are shared with administrators rather than only how a topic is configured. And the profile positions the holder as an implementer working beside architects, not as the architect.
What this credential does not prove
- Coding depth. The skills outline is about Power Fx, REST APIs and integration patterns. For Python and SDK-level agent work, Microsoft’s other agent exam is AI-103; see AB-620 versus AI-103.
- Administration. Tenant governance belongs to the administrators in the table above. The exam touches identity, security and governance as planning considerations, not as an admin certification. For the admin side of Copilot, look at AB-900.
- Currency. The credential lapses after a year unless renewed, so it shows recent contact with a fast-moving product. How that works is on the renewal and retake rules page.
- Production track record. Passing a proctored exam cannot show that an agent survived real users. Evaluation, test sets and lifecycle management are examined, but a portfolio of deployed agents says more.
Turning the duty list into a Copilot Studio portfolio
Microsoft’s duty list doubles as a checklist for a CV or portfolio. Each bullet is better backed by one concrete build than by the credential alone: an MCP server connected to an agent, a human-in-the-loop agent flow with error handling, a solution moved between environments with environment variables, a test set used to compare two versions. The test and manage domain covers the lifecycle items employers rarely see candidates describe well.
After AB-620: AI-103, AB-100, AB-900 or Agentforce
- Sideways, into code. AI-103 covers the code-first agent track on Foundry. The two exams are not a sequence, so the choice depends on the stack you want to work in.
- Upward, into design. AB-100 is the exam aimed at the architect roles that appear in the collaborator list. Its comparison with AB-620 shows how the scopes differ.
- Toward the administrators you work with. AB-900 is a fundamentals-level credential for the Microsoft 365 Copilot and agent administration side.
- Outside the Microsoft stack. Agentforce Specialist is the Salesforce equivalent for people whose employers run agents there.
For the wider picture across Microsoft’s AI credentials, see the Azure AI certification path.
Pay for Copilot Studio builders: where to look
This page quotes no salary or hiring figures for the credential, because none are published by Microsoft for it. For occupation-level numbers, use the U.S. Bureau of Labor Statistics Occupational Outlook Handbook or the equivalent statistics office in your country, and treat job adverts that ask for “Copilot Studio” by name as a better signal than any generic average.